Fix bug #307 : wrong NIC selection on the reduction tree.

The reduction tree (tree up) was inverting the NICs to use,
causing performance issue in cases where we are using different
NICs on a given channel.

src/include/graph.h

@@ -27,7 +27,7 @@ void ncclTopoFree(struct ncclTopoSystem* system);
 ncclResult_t ncclTopoTrimSystem(struct ncclTopoSystem* system, struct ncclComm* comm);
 
 // Query topology
-ncclResult_t ncclTopoGetNetDev(struct ncclTopoGraph* graph, int dir, int channelId, int* net);
+ncclResult_t ncclTopoGetNetDev(struct ncclTopoSystem* system, struct ncclTopoGraph* graph, int rank, int channelId, int* net);
 ncclResult_t ncclTopoCheckP2p(struct ncclTopoSystem* system, int64_t id1, int64_t id2, int* p2p);
 ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* topo, int64_t busId, int netDev, int read, int* useGdr);